Skip to content

Latest commit

 

History

History
80 lines (50 loc) · 2.99 KB

README-Generation-Uncond.md

File metadata and controls

80 lines (50 loc) · 2.99 KB

Unconditional Generation with GraphGPT Decoder

The generation of GraphsGPT is controlled by multiple adjustable configurations. You can refer to the following descriptions to adjust the generation to different needs.

Generation Configurations

  • Auto-fix Toggle:

    • strict_generation: (bool) True (Whether to tolerate the validity exceptions during generation. Setting to False will enable flexible generation that automatically fixes invalid predictions and ensure maximum effectiveness.)
    • fix_aromatic_bond: (bool) False (Whether to fix the dissociative aromatic bonds in the generated molecules.)

  • Sampling Strategy:

    • do_sample: (bool) False (Whether to use probabilistic sampling for bond predictions. Setting to True will enable probabilistic sampling and introduce more randomness.)
    • top_k: (int) None (The range of top predictions for probability sampling. Available when do_sample is True.)
    • temperature: (float) 1.0 (Temperature to adjust the probability distribution. Available when do_sample is True.)

  • Hyperparameters:

    • max_atoms: (int) None (The maximum number of atoms for generation.)
    • similarity_threshold: (float) 0.5 (Threshold for classifying whether a generated atom is new or old.)

  • Other Check Terms:

    • check_first_node: (bool) True (Whether to check the consistency between the predicted beginning atom and the first bond, and fix the order of the beginning two atoms.)
    • check_atom_valence: (bool) False (Whether to check the validity regarding the valence of the atoms connected to the predicted bonds.)

For reference, we provide some example configurations to use under different circumstances.

Validate the Pretraining Performance

To validate the pretraining performance, the generation should be of no randomness, both auto-fix and probabilistic sampling should be turned off:

strict_generation="True"
fix_aromatic_bond="False"

do_sample="False"

check_first_node="True"
check_atom_valence="False"

An example script can be found in scripts/generation/unconditional/examples/generate_strict.sh.

Generate with More Effectiveness

Upon generation of more effective results. You can turn on the auto-fix:

strict_generation="False"
fix_aromatic_bond="True"

do_sample="False"

check_first_node="True"
check_atom_valence="False"

An example script can be found in scripts/generation/unconditional/examples/generate_flexible.sh.

Generate with More Diversity (Further Finetuning Needed)

To generate more diverse results. You can further turn on the probabilistic sampling and valence check. This requires the encoded Graph Words $\mathcal{W}$ to be of variable information, where an extra finetuning would be needed.

strict_generation="False"
fix_aromatic_bond="True"

do_sample="True"
top_k=4
temperature=1.0

check_first_node="True"
check_atom_valence="True"

Visualize the Results

Run scripts/generation/unconditional/visualize.sh.